sungguk's picture
Release visual answerability benchmark v1.0.0
e1ced61 verified
Raw
History Blame Contribute Delete
1.7 kB
"""C2 independent dual-compile (``docs/02`` §7, ``docs/07`` P4).
Two independent models compile a canonical question program from the same
untouched evidence; their outputs are reconciled by canonical exact-match
(§7.2) and a mutation equivalence suite (§7.3). A C2 program is never used in
primary evaluation — the records feed training-time augmentation only.
* :mod:`state` — the five-state gate + outcome.
* :mod:`request` — the §7.1 independent compile request (hides intervention /
sibling output / transformed image).
* :mod:`client` — the gated model-client contract (never fakes a call).
* :mod:`parse` — model JSON → typed ``Program`` (one repair).
* :mod:`record` — schema-valid ``compiled_program`` emission.
* :mod:`mutations` — the §7.3 deterministic mutation suite.
* :mod:`compiler` — the orchestrator wiring the gate end-to-end.
"""
from __future__ import annotations
from .client import HfModelClient, ModelClient, ModelResponse, ModelUnavailable
from .compiler import C2Compiler
from .mutations import Mutation, deterministic_mutations
from .parse import ParsedProgram, parse_program_response
from .record import build_compiled_program
from .request import C2Candidate, CompileRequest, build_compile_request
from .state import C2Outcome, C2State, MutationEvidence
__all__ = [
"C2Candidate",
"C2Compiler",
"C2Outcome",
"C2State",
"CompileRequest",
"HfModelClient",
"ModelClient",
"ModelResponse",
"ModelUnavailable",
"Mutation",
"MutationEvidence",
"ParsedProgram",
"build_compiled_program",
"build_compile_request",
"deterministic_mutations",
"parse_program_response",
]